Re: [RFC PATCH v2] mm: vmscan: avoid anon scanning for GFP_NOIO with low swapcache in MGLRU

[Baolin Wang <baolin.wang@linux.alibaba.com>]

On 10/22/26 6:27 PM, Barry Song (Xiaomi) wrote:
> On Thu, Oct 22, 2026 at 4:49 PM Baolin Wang <baolin.wang@linux.alibaba.com> wrote:
>>
>>
>>
>> On 10/22/26 5:55 AM, Barry Song wrote:
>>> On Wed, Oct 21, 2026 at 3:28 PM Baolin Wang
>>> <baolin.wang@linux.alibaba.com> wrote:
>>> [...]
>>>
>>>>>     static bool lruvec_is_sizable(struct lruvec *lruvec, struct scan_control *sc)
>>>>>     {
>>>>>         unsigned long total;
>>>>>         int swappiness = get_swappiness(lruvec, sc);
>>>>>         struct mem_cgroup *memcg = lruvec_memcg(lruvec);
>>>>>
>>>>> +     swappiness = get_scan_swappiness(lruvec, sc, swappiness);
>>>>> +
>>>>>         total = lruvec_evictable_size(lruvec, swappiness);
>>>>>
>>>>>         /* whether the size is big enough to be helpful */
>>>>> @@ -5249,9 +5265,10 @@ static bool try_to_shrink_lruvec(struct lruvec *lruvec, struct scan_control *sc)
>>>>>         bool need_rotate = false, should_age = false;
>>>>>         long nr_batch, nr_to_scan;
>>>>>         int swappiness = get_swappiness(lruvec, sc);
>>>>> +     int scan_swappiness = get_scan_swappiness(lruvec, sc, swappiness);
>>>>
>>>> IMO, 'swappiness' and 'scan_swappiness' are really confusing. Why move
>>>> this logic into get_swappiness()? In get_swappiness(), we already have
>>>> some similar logic to adjust swappiness when there is no available swap
>>>> space.
>>>>
>>>
>>> Yes. It is very confusing. we can't do that in get_swappiness() as otherwise,
>>> we will get cold/hot inversion due to GFP_NOIO. for a real swappiness=0,
>>> we allow cold/hot inversion for anon, but for GFP_NOIO, this is not the
>>> case.
>>
>> But can_reclaim_anon_pages() doesn't only handle the GFP_NOIO case, it
>> also covers the swap space limitation scenario. The latter is what
>> concerns me more.
>>
>> If we're only talking about the GFP_NOIO case, I'd agree with your
>> concern. However, we could introduce a helper to force anon aging and
>> add comments explaining why we need this. That would look more
>> maintainable than 2 confusing swappiness variables, no?
>>
>> Just my 2 cents.
>>
>> Something like below (untested):
>>
>> diff --git a/mm/vmscan.c b/mm/vmscan.c
>> index daa7e37c2236..7c1d26e6b9d7 100644
>> --- a/mm/vmscan.c
>> +++ b/mm/vmscan.c
>> @@ -2890,7 +2890,7 @@ static int get_swappiness(struct lruvec *lruvec,
>> struct scan_control *sc)
>>           if (swappiness == SWAPPINESS_ANON_ONLY)
>>                   return swappiness;
>>
>> -       if (!sc->may_swap)
>> +       if (!sc->may_swap || !(sc->gfp_mask & __GFP_IO))
>>                   return 0;
> 
> For GFP_NOIO, we should also check whether swapcache is low, since
> GFP_NOIO can reclaim swapcache. We already have
> reclaimable_anon_is_low() for this.

Ah, yes, just a quick sample to show what I had in mind. :). Sould also 
consider reclaimable_anon_is_low() for the GFP_NOIO case.

>>           if (!can_demote(pgdat->node_id, sc, memcg) &&
>> @@ -4049,7 +4049,7 @@ static inline void flush_lru_batch(struct
>> list_head *head, struct list_head **ba
>>           *batch_end = NULL;
>>    }
>>
>> -static bool inc_min_seq(struct lruvec *lruvec, int type, int swappiness)
>> +static bool inc_min_seq(struct lruvec *lruvec, int type, int
>> swappiness, bool force_anon_aging)
>>    {
>>           int zone;
>>           int remaining = MAX_LRU_BATCH;
>> @@ -4063,7 +4063,7 @@ static bool inc_min_seq(struct lruvec *lruvec, int
>> type, int swappiness)
>>                   goto done;
>>
>>           /* For anon type, skip the check if swappiness is zero (file
>> only) */
>> -       if (!type && !swappiness)
>> +       if (!type && !swappiness && !force_anon_aging)
>>                   goto done;
>>
>>           VM_WARN_ON_ONCE(get_nr_gens(lruvec, type) != MAX_NR_GENS);
>> @@ -4183,7 +4183,20 @@ static void try_to_inc_min_seq(struct lruvec
>> *lruvec, int swappiness)
>>           }
>>    }
>>
>> -static bool inc_max_seq(struct lruvec *lruvec, unsigned long seq, int
>> swappiness)
>> +static bool is_force_anon_aging(struct scan_control *sc, struct lruvec
>> *lruvec,
>> +               int swappiness)
>> +{
>> +       if (swappiness || (sc->gfp_mask & __GFP_IO))
>> +               return false;
> 
> As above, GFP_NOIO doesn't imply swappiness=0 if there are many
> swapcache folios to reclaim. So we should check
> reclaimable_anon_is_low() here.
> 
>> +
>> +       /*
>> +        * Explain why we neeed to force increase anon min seq.
>> +        * xxxxxx
>> +        */
>> +       return swappiness != sc_swappiness(sc, lruvec_memcg(lruvec));
>> +}
>> +
>> +static bool inc_max_seq(struct lruvec *lruvec, unsigned long seq, int
>> swappiness, bool force_anon_aging)
>>    {
>>           bool success;
>>           int prev, next;
>> @@ -4205,7 +4218,7 @@ static bool inc_max_seq(struct lruvec *lruvec,
>> unsigned long seq, int swappiness
>>                   if (get_nr_gens(lruvec, type) != MAX_NR_GENS)
>>                           continue;
>>
>> -               if (inc_min_seq(lruvec, type, swappiness))
>> +               if (inc_min_seq(lruvec, type, swappiness, force_anon_aging))
>>                           continue;
>>
>>                   lruvec_unlock_irq(lruvec);
>> @@ -4249,7 +4262,7 @@ static bool inc_max_seq(struct lruvec *lruvec,
>> unsigned long seq, int swappiness
>>    }
>>
>>    static bool try_to_inc_max_seq(struct lruvec *lruvec, unsigned long seq,
>> -                              int swappiness, bool force_scan)
>> +                              int swappiness, bool force_scan, bool
>> force_anon_aging)
>>    {
>>           bool success;
>>           struct lru_gen_mm_walk *walk;
>> @@ -4260,7 +4273,7 @@ static bool try_to_inc_max_seq(struct lruvec
>> *lruvec, unsigned long seq,
>>           VM_WARN_ON_ONCE(seq > READ_ONCE(lrugen->max_seq));
>>
>>           if (!mm_state)
>> -               return inc_max_seq(lruvec, seq, swappiness);
>> +               return inc_max_seq(lruvec, seq, swappiness,
>> force_anon_aging);
>>
>>           /* see the comment in iterate_mm_list() */
>>           if (seq <= READ_ONCE(mm_state->seq))
>> @@ -4295,7 +4308,7 @@ static bool try_to_inc_max_seq(struct lruvec
>> *lruvec, unsigned long seq,
>>           } while (mm);
>>    done:
>>           if (success) {
>> -               success = inc_max_seq(lruvec, seq, swappiness);
>> +               success = inc_max_seq(lruvec, seq, swappiness,
>> force_anon_aging);
>>                   WARN_ON_ONCE(!success);
>>           }
>>
>> @@ -5262,7 +5275,9 @@ static bool try_to_shrink_lruvec(struct lruvec
>> *lruvec, struct scan_control *sc)
>>                   }
>>
>>                   if (should_run_aging(lruvec, max_seq, sc, swappiness)) {
>> -                       if (try_to_inc_max_seq(lruvec, max_seq,
>> swappiness, false))
> 
> This is quite tricky. If we pass swappiness=0, anon aging will be
> skipped. Do we want that for GFP_NOIO cases? I mean, these are not
> real cases where anon aging should be disabled.

No, we have real use cases where 'vm_swappiness' is set to 0, which 
means the workload never reclaims anon folios. So in this case we don't 
need to age anon folios.
> static int should_skip_vma(unsigned long start, unsigned long end, struct mm_walk *args)
> {
>          ...
>          if (vma == get_gate_vma(vma->vm_mm))
>                  return true;
>          ...
> }
> 
>> +                       bool force_anon_aging = is_force_anon_aging(sc,
>> lruvec, swappiness);
> 
> Adding a force_anon_aging argument to inc_max_seq() seems like
> quite a burden. Maybe a cleaner approach is to pass sc all the
> way to inc_max_seq() and let it detect GFP_NOIO cases itself?

OK, still look good to me.

> BTW, I guess we don't need to worry about swappiness at all. We can
> simply bypass anon reclaim?
> 
> diff --git a/mm/vmscan.c b/mm/vmscan.c
> index fc823fe3bd1f..fa6d9860156d 100644
> --- a/mm/vmscan.c
> +++ b/mm/vmscan.c
> @@ -4903,6 +4903,8 @@ static int scan_folios(unsigned long nr_to_scan, struct lruvec *lruvec,
>   	int skipped = 0;
>   	unsigned long remaining = nr_to_scan;
>   	struct lru_gen_folio *lrugen = &lruvec->lrugen;
> +	struct pglist_data *pgdat = lruvec_pgdat(lruvec);
> +	struct mem_cgroup *memcg = lruvec_memcg(lruvec);
>   
>   	VM_WARN_ON_ONCE(nr_to_scan > MAX_LRU_BATCH);
>   	VM_WARN_ON_ONCE(!list_empty(list));
> @@ -4910,6 +4912,9 @@ static int scan_folios(unsigned long nr_to_scan, struct lruvec *lruvec,
>   	if (get_nr_gens(lruvec, type) == MIN_NR_GENS)
>   		return 0;
>   
> +	if (!type && reclaimable_anon_is_low(memcg, pgdat->node_id, sc))
> +		return 0;
> +
>   	gen = lru_gen_from_seq(lrugen->min_seq[type]);
>   
>   	for (i = MAX_NR_ZONES; i > 0; i--) {
> @@ -5164,12 +5169,19 @@ static int evict_folios(unsigned long nr_to_scan, struct lruvec *lruvec,
>   static bool should_run_aging(struct lruvec *lruvec, unsigned long max_seq,
>   			     struct scan_control *sc, int swappiness)
>   {
> +	struct pglist_data *pgdat = lruvec_pgdat(lruvec);
> +	struct mem_cgroup *memcg = lruvec_memcg(lruvec);
>   	DEFINE_MIN_SEQ(lruvec);
>   
>   	/* have to run aging, since eviction is not possible anymore */
>   	if (evictable_min_seq(min_seq, swappiness) + MIN_NR_GENS > max_seq)
>   		return true;
>   
> +	/* run aging when GFP_NOIO limits anon reclaim with few swapcache folios */
> +	if (swappiness && min_seq[LRU_GEN_FILE] + MIN_NR_GENS > max_seq &&
> +	    reclaimable_anon_is_low(memcg, pgdat->node_id, sc))
> +		return true;

I've considered this, but we'd still need to add some code in 
get_nr_to_scan()/lruvec_is_sizable() to filter out the anon folios for 
GFP_NOIO case. Rather than spreading the logic across multiple places, 
I'd prefer adding a helper with proper comments to handle the aging 
issue, like the sample code I showed.